Always
It may be a simplistic view, but it seems to me that the answer to the 'when should I pass on PaaS' question is 'always'.
I see the role of SaaS for non-tech startups and non-core corporate applications where development time and focus are too expensive to waste when something can be pulled off the shelf. Lock-in is a significant worry, but the benefits can outweigh the risks provided the business won't come crashing to a halt if things go wrong.
I see the role of IaaS for companies that don't want to manage the hardware layer and for whom security and uptime are not so vital as to warrant the extra cost of a bricks and mortar or colo solution for smaller or variable workloads. This market will mature rapidly, and Amazon have shown that there's a significant appetite for it even though the long-term costs may be higher than inhouse for a lot of their growing customer base.
But I've never seen the role PaaS. It's a solution which still requires significant developer resource (including developer resource to tune the application to work with the quirks of the particular platform) but locks users tightly into an ecosystem that the provider can change at any time (as both Heroku and the Rackspace 'Cloud sites' service have both done to the chagrin of their user base).
PaaS offers all of the disadvantages of both forms of cloud while nullifying any tangible advantage. I don't see it having any long-term future. Some of the more useful proprietary features of Amazon may last for a while, but a few high profile examples of customers getting stung because they can't move away or the proprietary feature fails (as EBS appears to have done on a few occasions) will soon push things back to IaaS.