Think the problem is really down to a statistical assumption I've done. Basically Neutnet receives a message to spam, queries after bots and mains who should receive the message, and now some of the somewhat flawed logic starts. It starts to loop around all the receivers, and the receiver is assigned to a satellite which has the lowest countindex. If the first receiver has 9 alts, the spambot sets the countindex 10 so at least the 19 next messages is sent through Neutnet2-20. The satellites now checks if the main is online, and if it's not, the 9 alts is checked individuallly. This algorithm works pretty well when you look at how many potential charactesr a satellite checks if is online. The difference between highest and lowest countindex is usually <1% of total people potentially checked. What the satellite currently lacks though, is some counter that indicates how many messages it has actually sent...and there is the problem I think. Statistically it should in the end really even out between the spambots, but it looks like at peaktimes some satellites send out a significant amount of messages more than other bots, thereby slowing messages for people that is supposed to get some message from said satellites. It's not the computer or anything like that...it's not particularly intensive on the CPU or anything, it's basically statistics working against us (and some crap coding)
Increasing number of satellites isn't really going to work all that much at the moment. I really need to work on the pipeline for how messages are sent to satellites.