Validators only request one collation at a time per relay parent and collators only send one collation at a time per relay parent. The problem with that is that if there is a slow validator/collator it could lead to a stall of the parachain. This is also a potential DoS vector as the receiving/sending node could send the data on purpose rather slow. To prevent this, we should introduce a second timeout. This timeout should be chosen wisely and should reflect the time we expect the other side to have received/send the data. If this timeout is hit, we should start requesting a second collation/start sending our collation to a second validator.

This approach assumes that if someone isn't fast enough to send/receive the collator in a reasonable amount of time, it also doesn't uses our full bandwidth and we can start to share the bandwidth with others.