Swahili Gender Agreement

Okay, so I've doing some more swahili recently.. and more and more I've
been wondering how the gender system actually survives in its current
form. With the amount of agreement there is in a sentence, with
adjectives agreeing in class (and thus in number) with their nouns, the
verb agreeing with both the subject and object and also having
compulsory indication of tense/aspect and mood (well, in bantu verbs,
the -a at the end of the dictionary form is an indicative mood marker,
although some verbs borrowed from arabic don't change to indicate mood)
and all of these being separate morphemes glued together instead of
merging... It seems to me that the need to be concise would have led to
either some of the affixes fusing, or just being dropped, a long time
ago, especially since word order in swahili is generally fixed so the
agreement seems to serve little purpose. Its very interesting just... I
would have thought that the need to state things in the most concise way
possible (allowing for sufficient redundancy to make communication
reliable) would be one of the primary driving forces of language
evolution. The one thing that the gender system seems to have going for
it is the fact you can use it to derive new words sometimes, ie:

m-toto
PERSON-child            child
u-toto
ABSTRACT-child         childhood

And so on.
Anyway... All I'm saying is, I find the structure of the language
fascinating, but I can't understand how that structure is stable. It
seems to be, since the classes and the extreme level of agreement
doesn't seem to have been reduced much since proto-Bantu down to the
modern Bantu languages, but there is so much agreement which is rendered
useless by quite strict word ordering in Swahili it seems strange that
it isn't in the process of being reduced.

Chris.