<Tracemonkey Big Picture>
                            
I tried on Tracemonkey big picture.

In summary, XPCOM calls public APIs when parser of Gecko meets javascript block (<script> ~ </script>).  Then, the javascript is parsed by parser of Tracemonkey. The parsed javascript is emited to bytecode for interpreting.
The bytecode is processed by interpreter of Tracemonkey. In the Tracemonkey, interpreter is in charge of the hot spots detection. To my knowledge, the Tracemoneky recognizes loop as hot spots. However, I don't know how much the algorithm is improved so far. I believe the hot spot detection is being improved by Brendan Eich, Andreas Gal and so on. Anyway, the hot spots is converted to native code by JITc(Just-in-time compiler) in order to process more faster than bytecode interpretation.

Lastly, Garbage Collelector manages allocation of memory in Tracemonkey, However, the Garbage collector is using mark and sweep algorithm. it is very primitive algorithm. It should be changed with further improved algorithm of Garbage Collection. For example, generation, conservative GC algorithm.

Recently, Safari 4 beta. was released. The performance is very surprise !!. According to the announce, Safari 4 uses Nitro JavaScript Engine, It seems to me that the Nitro JavaScript Engine is renamed from Squirrelfish Extreame. I look forward to seeing more faster Firefox 3.1 via Tracemonkey.