Blockchain reactor and blockpool parameters need some better tuning:

• maxPendingRequestsPerPeer should be set based on how many blocks we could be sending to a peer at any one time. Eg. for a 128kB/s connection with 1kB blocks and 10 peers, we can push 12.8 blocks/sec per per.
• requests and blocks share the same reactor channel - this should be split in two
• we use TrySend everywhere, meaning if our reactor channel is full, we'll drop the msg and likely timeout

In general the pool is a bit complicated with its use of maps/mutexes and could use some cleanup or at least make it clearer whats happening where and why