Unfortunately the question is, strictly speaking, unanswerable.
The best one could do is come up with a specific design and estimate the number of gates needed, and then one couldn't be sure exactly how suboptimal the design in question is, regarding gate count.
One of the most basic design choices is probably how much and which work to do using dedicated logic, and how much to do using general purpose logic and microcode.
One may get an idea of the order of magnitude of gates required by taking a minimal CPU core (say, the j1 core from <a href="http://excamera.com/sphinx/fpga-j1.html" rel="nofollow">http://excamera.com/sphinx/fpga-j1.html</a>), writing a minimal lisp interpreter on that core, and adding the gates needed to store the interpreter to the number of gates used for the core. This is left as an exercise for the reader.
To figure that out, you would have to try it.<p>I guess you Googled and found some interesting things like;<p><a href="http://www.frank-buss.de/lispcpu/" rel="nofollow">http://www.frank-buss.de/lispcpu/</a><p>but no answer to your question. So get out your pen & paper and build one :)
<a href="http://lambda-the-ultimate.org/node/3583" rel="nofollow">http://lambda-the-ultimate.org/node/3583</a> might be a good starting point assuming that an implementation of lambda calculus qualifies as a LISP (it should).