diff --git a/ormar/models/model_row.py b/ormar/models/model_row.py
index 41ddb2074..1110b2f29 100644
--- a/ormar/models/model_row.py
+++ b/ormar/models/model_row.py
@@ -1,4 +1,5 @@
 from dataclasses import dataclass
+from functools import lru_cache
 from typing import TYPE_CHECKING, Any, Optional, Union, cast
 
 try:
@@ -39,6 +40,40 @@ class RowExtractionPlan:
 PlanCache = dict[tuple[type, str, int], RowExtractionPlan]
 
 
+@lru_cache(maxsize=2048)
+def build_shared_row_plan(
+    model_cls: type["ModelRow"],
+    table_prefix: str,
+    include: frozenset,
+    exclude: frozenset,
+    pk_only: bool,
+) -> RowExtractionPlan:
+    """
+    Builds a row extraction plan shared between querysets. A plan only depends
+    on the model, the table prefix and the model's own include/exclude/pk_only
+    settings, so those are the cache key.
+
+    :param model_cls: model the plan is built for
+    :type model_cls: type[ModelRow]
+    :param table_prefix: prefix of the table from AliasManager
+    :type table_prefix: str
+    :param include: fields to include for this model
+    :type include: frozenset
+    :param exclude: fields to exclude for this model
+    :type exclude: frozenset
+    :param pk_only: flag if only the primary key is selected
+    :type pk_only: bool
+    :return: extraction plan for this row position
+    :rtype: RowExtractionPlan
+    """
+    excludable = ExcludableItems()
+    model_excludable = excludable.get(model_cls=model_cls, alias=table_prefix)  # type: ignore
+    model_excludable.include = set(include)
+    model_excludable.exclude = set(exclude)
+    model_excludable.pk_only = pk_only
+    return model_cls.build_row_extraction_plan(table_prefix, excludable)
+
+
 class ModelRow(NewBaseModel):
     @classmethod
     def from_row(  # noqa: CFQ002
@@ -417,9 +452,9 @@ def get_or_build_row_plan(
         plan_cache: Optional[PlanCache],
     ) -> RowExtractionPlan:
         """
-        Return a cached plan for the given key, or build and cache one. When
-        ``plan_cache`` is ``None`` (legacy / external caller) the plan is
-        built fresh on every call so behavior matches the pre-cache shape.
+        Return a plan from the per-queryset cache, falling back to the plans
+        shared between querysets. When ``plan_cache`` is ``None`` (legacy /
+        external caller) only the shared plans are used.
 
         :param table_prefix: prefix of the table from AliasManager
         :type table_prefix: str
@@ -432,14 +467,38 @@ def get_or_build_row_plan(
         :rtype: RowExtractionPlan
         """
         if plan_cache is None:
-            return cls.build_row_extraction_plan(table_prefix, excludable)
+            return cls.get_shared_row_plan(table_prefix, excludable)
         key = (cls, table_prefix, id(excludable))
         plan = plan_cache.get(key)
         if plan is None:
-            plan = cls.build_row_extraction_plan(table_prefix, excludable)
+            plan = cls.get_shared_row_plan(table_prefix, excludable)
             plan_cache[key] = plan
         return plan
 
+    @classmethod
+    def get_shared_row_plan(
+        cls, table_prefix: str, excludable: ExcludableItems
+    ) -> RowExtractionPlan:
+        """
+        Return a plan shared between querysets for the model's own
+        include/exclude/pk_only settings at the given table prefix.
+
+        :param table_prefix: prefix of the table from AliasManager
+        :type table_prefix: str
+        :param excludable: structure of fields to include and exclude
+        :type excludable: ExcludableItems
+        :return: extraction plan for this row position
+        :rtype: RowExtractionPlan
+        """
+        model_excludable = excludable.get(model_cls=cls, alias=table_prefix)  # type: ignore
+        return build_shared_row_plan(
+            cls,
+            table_prefix,
+            frozenset(model_excludable.include),
+            frozenset(model_excludable.exclude),
+            model_excludable.pk_only,
+        )
+
     @staticmethod
     def apply_row_plan(
         plan: RowExtractionPlan,
diff --git a/ormar/queryset/actions/order_action.py b/ormar/queryset/actions/order_action.py
index 7e9506e87..f0d498e3c 100644
--- a/ormar/queryset/actions/order_action.py
+++ b/ormar/queryset/actions/order_action.py
@@ -1,4 +1,5 @@
 import copy
+from functools import lru_cache
 from typing import TYPE_CHECKING, Optional
 
 import sqlalchemy
@@ -10,6 +11,20 @@
     from ormar import Model
 
 
+@lru_cache(maxsize=2048)
+def order_text_clause(expression: str) -> sqlalchemy.sql.expression.TextClause:
+    """
+    Builds a text clause for an ORDER BY expression, cached as text clauses are
+    immutable and the same orderings (e.g. default pk sort) repeat on every query.
+
+    :param expression: raw ORDER BY expression
+    :type expression: str
+    :return: text clause for the expression
+    :rtype: sqlalchemy.sql.expression.TextClause
+    """
+    return text(expression)
+
+
 class OrderAction(QueryAction):
     """
     Represents a single ORDER BY directive on a queryset.
@@ -94,7 +109,9 @@ def get_text_clause(self) -> sqlalchemy.sql.expression.TextClause:
         else:
             table_name = quoter(f"{prefix}{table_name}")
         field_name = quoter(field_name)
-        return text(self._build_order_expression(f"{table_name}.{field_name}"))
+        return order_text_clause(
+            self._build_order_expression(f"{table_name}.{field_name}")
+        )
 
     def _build_order_expression(self, full_column: str) -> str:
         """
diff --git a/ormar/relations/alias_manager.py b/ormar/relations/alias_manager.py
index d1ea622b3..33944196a 100644
--- a/ormar/relations/alias_manager.py
+++ b/ormar/relations/alias_manager.py
@@ -1,5 +1,6 @@
 import string
 import uuid
+from functools import lru_cache
 from random import choices
 from typing import TYPE_CHECKING, Any, Optional, Union
 
@@ -13,6 +14,37 @@
     from ormar.models import ModelRow
 
 
+@lru_cache(maxsize=2048)
+def build_prefixed_columns(
+    alias: str, table: sqlalchemy.Table, fields: Optional[frozenset[str]]
+) -> tuple[Label[Any], ...]:
+    """
+    Builds labelled columns of a table, cached as labels are immutable and the
+    same model/alias/fields combination is selected over and over.
+
+    :param alias: alias of given table
+    :type alias: str
+    :param table: table from which fields should be aliased
+    :type table: sqlalchemy.Table
+    :param fields: column names (ormar aliases) to include, None for all
+    :type fields: Optional[frozenset[str]]
+    :return: tuple of labelled columns
+    :rtype: tuple[Label[Any], ...]
+    """
+    alias = f"{alias}_" if alias else ""
+    aliased_fields = {f"{alias}{x}" for x in fields} if fields else set()
+    all_columns = (
+        table.columns
+        if not fields
+        else [
+            col
+            for col in table.columns
+            if col.name in fields or col.name in aliased_fields
+        ]
+    )
+    return tuple(column.label(f"{alias}{column.name}") for column in all_columns)
+
+
 def get_table_alias() -> str:
     """
     Creates a random string that is used to alias tables in joins.
@@ -78,18 +110,9 @@ def prefixed_columns(
         :return: list of sqlalchemy text clauses with "column name as aliased name"
         :rtype: list[text]
         """
-        alias = f"{alias}_" if alias else ""
-        aliased_fields = [f"{alias}{x}" for x in fields] if fields else []
-        all_columns = (
-            table.columns
-            if not fields
-            else [
-                col
-                for col in table.columns
-                if col.name in fields or col.name in aliased_fields
-            ]
+        return list(
+            build_prefixed_columns(alias, table, frozenset(fields) if fields else None)
         )
-        return [column.label(f"{alias}{column.name}") for column in all_columns]
 
     def prefixed_table_name(
         self, alias: str, table: sqlalchemy.Table