Comandante77, I fixed this bug in the 1mb_hardforked, lowmem, and 1mb_segwit branches on my github. The main commit that addressed this issue is this one:
https://github.com/jtoomim/p2pool/commit/428bbe4211c39fe3175a9464fb253a699b792e5eThe problem is that p2pool has to guess what a reasonable pseudoshare difficulty is when it doesn't know how much hashrate the node has (i.e. on startup), and if you are running a node with a lot of hashrate, it can guess terribly wrong. If the difficulty is too low, your node will get flooded with shares, and the hashrate that your node sees will be limited by your node's CPU power in processing shares instead of being limited by your miners' actual hashrate. This makes your node think its hashrate is low, which makes it lower the pseudoshare difficulty, which makes the problem worse.
I fixed it by setting the initial pseudoshare difficulty to equal 1/100th or 1/1000th of the p2pool minimum real share difficulty, which results in about 1 share every 30ms to 300ms if you have the full pool on your node (i.e. you're solo mining).