I don't know of any good performance benchmark for Emacs, sadly.
Part of the reason is that it depends on what you want to measure (redisplay performance? I/O performance? ...).
Regarding Elisp performance in particular, the lack of benchmark is a symptom of the relative lack of work on improving its performance. FWIW, when working on the implementation of lexical-binding, I did use a benchmark to try and make sure that this new feature did not come at the cost of lower performance.
This benchmark I used is:
rm **/*.elc lisp/loaddefs.el; make -jN
running it from within Emacs's source causes all Elisp files to be recompiled, and the byte-compiler is one of the relatively few pieces of Elisp which run for long periods of time and can easily be run in batch mode.
The downside is that this is a moving target, since the time taken depends on the version of the byte-compiler you're using as well as on the version of Emacs you're compiling.