cannot afford

It sounds like that ESX's network performance is not good at all.
The more weird thing is that when running network sending from one guest, the specweb clients are all broken. And the first time that I can catch a chance to evaluate the transmiting rate which is around 68Mbps.

first time, all four down because kernel network panic.
Second time, two broken, complaining exceeded max allowed overthink time of 20 sec. Please make sure that neither the server or client are overloaded. If server is overloaded, consider the nunber of SIMULTANEOUS_SESSIONS requested. If clients appear overloaded, add more clients. The rest two: one 6.71%good, 16.78%tolerant, Total errors: 66; the other 0%good, 0%tolerant, Total error: 207.

I ping google, it has sent 14 packets, 6 received, the packet lost 57% , the response time is 19.0ms.

In ESX, the support sessions are limited to 20, when with 5 sessions, it still have some degradation here. But when fewer guests running there, the degradation will be relieved in some extent.

The side effect is that when the network goes wrong, the client will also be broken. So it is hard to do anything to solve the problem.

The kernel also reports a network panic there. Lost Internet connections

All these may due to:
1. ESX performance overhead is too high
2. Bladecenter network card or infrastructure isn't well designed
3. Network driver is not well written
4. The other blade network usage has affected the testbed.

-------------------------------------------------------------------
Summary
guest sending packets 4 broken, none suvives
guest receving packets the misbehaved broken at the 3rd iteration, the degradation is 60%, the good ones survive with 40% degradations.
The receiving rate is 10Mbps
ping google while guest receiving packets, 7packets , 5 received, 28% packet loss, the response time is 20 ms /40ms