I will take a risk and advise to redefine requirements. (I hope you do not hate me for it)
With BGP you should take advantage of what it offers. Instead of just load sharing
which is IGP thinking you can find out from those ISPs which prefixes are best (easiest, fastest to reach) for each one. Then create communities for each ISP, for example community 1,2,3 to group them. After that on each router of your customers, configure route maps that will match those communites and accordingly modify local preference. That way each customer will take the best path for those routes. That way you will prevent for example from situation packets of going to ISP1 to reach ISP3 and use ISP2 as a transit. That could happen if you lock each customer with specific ISP.
Of course if any ISP fails your BGP will failover to remaining.
Help here:
http://www.cisco.com/en/US/tech/tk365/technologies_configuration_example09186a00801475b2.shtml
Eventually you can do load sharing:
http://www.cisco.com/en/US/tech/tk365/technologies_configuration_example09186a00800945bf.shtml#conf4
It is based on chaning local preference based on access lists.
In this case it is easier to implement, but you pretty much slicing internet routes into slices and give to customers. However you have no clue if those slices will be even "thicknes". This is of course abstraction. Another words you may match less or more using access lists from that example. Therefore your loadsharing be utilize certain ISP more or less with significant difference. Good thing about this solution is you do not need to change ISPs configs and in the future it will stay like it.
