Sure I can understand its a fair amount of data, And that it might be sketchy software/ Poor server hardware.
This. It's unlikely to be a software issue because a) it works most of the time and b) he has said the database needs "catching up". This is a classic indicator of database I/O starvation. H20 is using AWS and for a database of this nature, you need high I/O. Amazon offers EBS storage blocks, but these have terribly variable performance and low IOPS. They did create two new SSD instances types (HI1, I2) but those are all sold out. So he's probably struggling with dropped database connections, long running queries, dropping and re-adding EBS volumes to his software raid 10, etc.
The coin trading algo is most likely database driven as well, and very time-sensitive, so my concern is that coin trading efficiency suffers when this happens as well.