My understanding, that some FTP transfers are far below the allocated bandwidth of their connection. They approach full bandwidth
on FTP transfers that are "in-country", but this degrades when they FTP from remote sites (e.g., U.S. FTP sites).
I can't overemphasize the role of delay in the TCP windowing algorithm, which( if we assume no packet loss nor a bottleneck in Maximum Segment Size), is the most likely factor that is leading to the FTP throughput reductions.
TCP ,without RFC1323 extensions, can set receive buffer window upto a maximum 64 Kbytes. This is adequate for most LAN implementations.
FTP'ng from the US to Panama is a classic "Long Fat Network" (LFN) scenario, in which TCP throughput is highly affected by the long RTT's.
e.g., Assuming a buffer of 50Kbytes, and 20 ms RTT, *Note: optimal buffer size is considered to be 2 x BDP, where BDP is the Bandwidth*Delay Product, so for a 20 ms RTT, and a 10 Mbps connection, that would be 2 x (10Mbps/8 * .020s) = 50Kbytes
(50K x 8)/.02 = 20 Mbps
Which is more than adequate for a 10 Mbps service, but insufficient for a 100 Mbps service.
If you use an FTP session from Chile to Seattle(US), you will experience a RTT of at least 170 ms. In this scenario, the maximum TCP throughput would be:
(64K x 8)/.170 = 3 Mbps.
You can see the deficiency with the greater RTT.