[c-nsp] BGP to OSPF redistribution

[null zeroroute]

I'm having a problem trying to figure out a way to get eBGP learned routes
(from a layer-3 VPN MPLS WAN provider) into our internal OSPF, so that the
routes learned via the provider are preffered over the internally learned
OSPF routes.

No matter where the BGP-->OSPF redistribution point is, if it's the PE or
CE, the routes will still show up (by default) as OSPF external, and will
never be prefferred.

The provider who's path we prefer will only run BGP.  We would like to use
OSPF everywhere if possible, for several reasons.

WAN provider A is a layer-3 VPN MPLS network, and is the prefferred path.
WAN provider B is a layer-2 VPN MPLS network over which we run OSPF.
Provider B's network is inferior at times and we use it as a backup.

The equipment where the eBGP peering relationsips exist is a mix of 7600,
3800, 2800, 1800, 6500, 3750, 3550.

We considered GRE over the providers network however we then wind up with
25+ tunnels at each location, and that just grows as each new site is added,
not to mention some potential issues regarding throughput with a GRE tunnel
in the path.

Is there a way to redistribute BGP into OSPF so that the routes can be
anything but OSPF external?

I have not found a way to do this yet, and was wondering if it's even
possible, or if I'm missing something obvious.  Any suggestions appreciated.