Your PingPlotter graph looks like classic bandwidth saturation. Upload bandwidth is usually a lot more limited than download bandwidth, and the symptoms I see in the graph are consistent with saturation.
The final destination, though, doesn't seem to be exhibiting the same problems - you turned off the time graph for the final destination, so it's hard to comment on that with too much authority, but it looks like maybe the packet loss and latency at the intermediate hops is a lot less than at the final destination. This indicates that an early hop is doing some prioritization of packets and TTL expired packets (intermediate hops in PingPlotter) might be getting a lower priority than FTP packets.
It's hard to know about the disconnect. If your FTP software is very latency sensitive, then I can imagine an uploading causing problems. FTP software, though, should not be very latency sensitive - just because the FTP protocol is so often used in situations where the bandwidth is saturated (since uploading and downloading at a maximum rate is so important for most users).
If you can easily reproduce the dropped sessions, I'd try to show your ISP how to do that - they may be able to do interesting things in solving the problem if it's easy for them to reproduce.