Hi Vikram,
Well, look at it this way, in the case of MPLS VPN, BGP between the PEs (address-family ipv4 vrf x) is used to transport the customer routes between CEs, and thus it is very logical that it should redistribute all kind of routes (in the case of MPLS VPN the customer routes are always going to be iBGP anyway since they are redistributed from IGP (PE-CE) on another PE - except on the local router when the customer is using BGP as the PE-CE routing protocol) - using another prospective, redistributing iBGP routes into IGPs may cause routing loops to form within an autonomous system, but in the case of PE-CE the redistribution is done into a routing protocol outside the BGP domain (customer side).
I hope that i've been informative.
BR,
Mohammed Mahmoud.