Opty is still better, since you can overclock hugely, even without speacial cooling.
The opterons as I understand it do have 128kbs per core, just that it is often listed wrong because the cache is divided into two 64kb sections per core (data/instructions). I could be wrong but thats just what Ive read/heard.
The ocing potential would totally waste the performance gain from the extra 64KB of L1 cache.