I think it comes down to history. Host names existed before domain names. When domains were bolted on they used the idea of a default domain for each host and that made sense to be on the end.
Consider:
telnet hosta # established way
telnet hosta.abc # domain bolted on back
telnet abc.hosta # domain bolted on front
Since people knew the host names and were used to dealing with them, the suffix was more natural since it kept the domain cruft out at the edge.
This seems pretty much right based on the historical data.
It doesn't seem like ordering of domain name parts was given much thought. RFC 882, which first defined the domain name space, said only this:
By convention, the labels that compose a domain name are read left to right, from the most specific (lowest) to the least specific (highest).
Additionally, it seems that when the first TLD was defined, .arpa, some people were hacking to support it by just concatenating .arpa to the end of all the domain names. Concatenation, of course, is much easier than prepending when you are programming in C. Note the following from RFC 881, which described the transition to using domain names, at a time when the domain name mapping was all stored in HOSTS.TXT files:
So far, no new domains have been introduced. Only a table with all the entries having official names in the ARPA domain has been provided. This should allow programs to be constructed to deal with domain style names in a general way without any special hacks to add or delete the string ".ARPA" to or from host names.
Consider:
Since people knew the host names and were used to dealing with them, the suffix was more natural since it kept the domain cruft out at the edge.