[cogman10]

I've not seen this in other responses, but an answer to your question is every logical register adds overhead to context switching by the operating system.

The OS has to store and load all registers whenever it decides to switch which thread is processing. 100 more logical registers means 100 more locations the OS has to keep track of.

This is part of the reason why new SIMD instruction sets need OS support before you can start using them.