There are two obvious optimizations possible in the code that handles installation of a new topology.
1. Currently a new topology is not confirmed until the node successfully sends START_STATE_TRANSFER requests to all nodes it wants to fetch segments from. This does not wait for the actual data to arrive but it still blocks quite a lot. To fix this we need to confirm the topology right after transactions were received.

2. Fetching transactions from other members is done serially. To improve it we could split it into multiple concurrent tasks.