[c-nsp] Design Q

[info at beprojects.com]

I've got a design question that I am not sure about and was wondering if 
anybody has any insight.

My intention is to have two separate Internet routers running eBGP to 
two separate upstreams and then iBGP to each other.  On the LAN side, 
they will each have their own IP, say .2 and .3 and share a HSRP 
address, say .1.  If I'm on the LAN, my DG is .1.  If I try to go to 
some random Internet address, I will go to .1 and we'll say that right 
now, that is router A.  If the address is best reached through it, it 
will forward the packet.  If it is best reached through B, which it 
learns via iBGP, it will send an ICMP redirect, assuming they are 
enabled, and tell me to go to .3 instead, which will forward the packet. 
  So on my pc, I will build a routing table as I move along.

The question is, what do I do if I want to create 4-5 VLAN's.  Say I 
have a layer3 switch like a 3750.  My initial thought would be to make 
it the DG for each vlan and it would have a default route of .1. 
However every time it gets a redirect, it will add it to it's routing 
table, right?  Now if one of the machines on one of the vlan's is a dns 
server or web server, theoretically, there could be thousands on 
different destinations, in which case the L3 switch will have thousands 
of routing entries and we all know that a lower end switch like the 3750 
does not have enough memory to hold 100,000 routes.

Am I better off setting up the links to the routers as trunks and using 
sub interfaces?  Or am I way off in my description above?  Or is there 
some other option?

Thanks.

Peder