[perf] esp adjust hardcoded-delta fast handlers

_From [derek.br...@gmail.com](https://code.google.com/u/117968039472581148324/) on December 10, 2010 18:06:10_

PR 613431

perlbmk and gcc still have significant overhead and sampling shows a lot of it is in esp adjust handling

this case covers trying to have some hardcoded fast loops for common small sizes. the dispatch on size
should be less overhead than always using the general loop that handles all sizes.

initial experiments showed that gcc does have a slight win from this approach.

_Original issue: http://code.google.com/p/drmemory/issues/detail?id=208_