Hello Jerry,

> Site 1                                      Site 2
> ------                                      ------
> MTA gateway                                 MTA gateway
> (and coda client)                           (and coda client)
>    |                                           |
>    |                                           |
> MTA server                                  MTA server
> (coda client     / coda volume(s) for \     (coda client
>    and           |                    |      and
> coda server)     \ maildir storage    /     coda server)

(unclear for me why you need a Coda client on both MTA gateway and MTA server?
Which protocol is used between them?)

I can not answer your main question, but want to provide some input.

One misconception about Coda is that it is optimized for
distantly placed servers, it is not.

When clients become weakly connected (which means talk to just one of
the servers) - and this is the normal mode of operation,
there will be more traffic between the servers than between the clients
and the servers.

If your line can bear with the excessive traffic, you are fine,
and you are protected against one server going down for short
periods of time, or the line breaking - again, for
short periods of time. You can not afford running longer in "reduced mode",
the server logs will eventually fill up and you will loose consistency.

> Q1: Can T1 speed adequately support two Coda servers?

Is it adequate for at least twice your mail traffic?

> Current size of the maildir's is 21 GB, but growth is expected.

Be aware that you cannot have a single directory bigger than several thousand
entries (directory size is limited to 256K I think).
It is probably the biggest concern for maildir-based installations.

Be also aware about server file number limit. 21G data should be ok,
twice as much (in small files) will be probably too much if you do not
distribute data between several servers.

> Q2: What are practical size limitations of Coda volumes?

According to Jan there shouldn't be any, though you probably do not want
to have huge volumes. I think at least some operations on a volume are
serialized, so the bigger volumes, the bigger risk for delays when client
processes wait for each other.

Hope it helps,
--
Ivan