there are links to several 8052 softcores in the Hamburg VHDL archive. Also on http://www.opencores.org

the best one to use depends on what your application calls for,the ideal being one which you can easily modify for your requirements in terms of the number of timers/interrupts/hardware multipliers etc needed.Clearly if your application doesnt use instructions which use decimal adjust/multiply/divide then you save a lot of hardware resorces by removing the hardware modules from the core.As I wrote in the FAQ support of open source cores is strictly limited and if you modify a core its upto you to verify that it still works.
one of the bigest bottle necks you'll find is the hardware multiplier,if your FPGA doesnt have mutilpliers built in then the harware compiler will produce one from its library of macros and there is a trade off betwen size of the logic and speed of execution.You can have one which will multiply in a single clock cycle and be huge in terms of the logic required ,or small but require multiple clock cycles.