Wouldn't matching a bit and excluding by another bit still be much cheaper?
The best case is a simple mask. I don't see how that can be done, given how Unicode can reclassify if something is a whitespace.
The worst case is 25 if-equals tests, so I'll call it 25 cycles since most text is not whitespace.
A 32-bit multiplication, shift, an L1 lookup, and comparison is fast. Eyeballing I'll guess about 8 cycles.
A mask and compare would be, what, 2 cycles?
While the loop to read these code points -- likely with utf-8 decoding -- adds its own overhead, plus the function call overhead, of course.
I am very much not a low-level developer, so take these numbers with caution. They are meant only to give some estimate on what "much" means.