I played ~30 matches with this, found no issues. I tested by looking at the dead bodies closely when the owner of the dead body respawns. They still do twitch sometimes, but this also happens on vanilla, i.e. I'm pretty sure that this doesn't have to do with this MR.

How did you measure the reduced network data if I may ask?
If the value is different from what it was without setting these to 0 you're still actually sending the delta difference which may not actually save anything but end up transmitting more.

Oops, you're right. I wanted to clarify this, but forgor 💀
I edited the description.

If the value is different from what it was without setting these to 0 you're still actually sending the delta difference which may not actually save anything but end up transmitting more.

According to the output of cl_showned, the zeroed struct is used as a basis for delta. I suspect that this has to do with the fact that the entity is initially unlinked.

I also initially wanted to go for minimizing changes, but then found this out.

the zeroed struct is used as a basis for delta

OK, I also double-checked that this is not just a logging error. With cl_shownet 3 it prints baseline as opposed to delta.
https://github.com/ec-/Quake3e/blob/a0e6451e22ea70e5530483863a6018c2d2a5c79d/code/client/cl_parse.c#L141-L167

Baselines are initialized here. Only the entity number is set there. But regardless of what the baseline is, angles and eventParm are dynamic anyway, so even if baseline is not 0 then sometimes we will still send the delta.

And one can look at just the message size. It's 45 vs 41.

(After discovering the concept of "baselines" I went into the rabbit hole and made #81. Thanks @ensiform!)