I have sysload 20-40% on shard.
polcore().error_creations_per_min return 105189
Its normal? How i can find scripts who create errors?

well, you could check debug.log (if you have that option enabled in pol.cfg), it will probably tell you where at least most of the errors are. Thats probably not entirely whats causing the lag though.
Whats your runaway script threshold?
Do you have lots of scripts that just say sleepms(1); just to stop it from running away?
How many connections to your shard?
How many script instructions/min on your shard?
What type of specs are on your server? (Processor speed, amount of RAM, RAM speed, etc...)