bitcoind JSON-RPC interface timeouts under unit tests

Question

I am running unit tests against bitcoind which do JSON-RPC requests in high rate. Random failures are an issue. E.g. if you have something like this in printconsole logs:

ThreadRPCServer method=gettransaction
ThreadRPCServer method=gettransaction
ThreadRPCServer method=gettransaction
ThreadRPCServer method=gettransaction
....
ThreadRPCServer method=gettransaction
ThreadRPCServer method=gettransaction
sending: inv (37 bytes)
ThreadRPCServer method=gettransaction
ThreadRPCServer method=listtransactions
sending: inv (37 bytes)

You start getting:

synced with network?".format(name)) from e
cryptoassets.core.backend.bitcoind.BitcoindJSONError: Got timeout when doing bitcoin RPC call listtransactions. 

I have tried with both reusing HTTP connection and not reusing HTTP connection with various results.

Are there any low hanging fruits to make bitcoind RPC interface more stable (increase threads, concurrency, etc.?)

I am using the following Python JSON-RPC implementation:

https://github.com/jgarzik/python-bitcoinrpc/blob/master/bitcoinrpc/authproxy.py

UPDATE: Looks like the problematic call might be listtransactions. When called at the same the time from command line using curl there is no response either:

[~]% curl --user foo:bar --data-binary '{"id":"t0", "method": "listtransactions", "params": ["cryptoassets"] }' http://127.0.0.1:8332/ |python -m json.tool
  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
  0     0    0     0    0     0      0      0 --:--:-- --:--:--100    70    0     0  100    70      0     58  

Answer 1

I used BitcoinLib to perform 15,000 consecutive listtransactions calls and the result was:

0 timeouts
Fastest response: 0.1133479 sec
Slowest response: 0.6475277 sec

bitcoind was started with the default parameters and was receiving blocks during the test.