I read with interest Phil's post on SaaS appliance and how a few companies are using this model for "putting SaaS on-premise" as Phil says. Phil's commentary is very similar to what we had published in the past in our enterprise perspective white paper here, or Fred's post here where software as a service can be seen as a continuum of possibilities along multiple axis: location being one of them. Other continuum can be around licensing, life cycle management, UX etc.
Coming back to the appliance model, one thing that is important to note is that the appliance model is very likely to use an isolated model (1 appliance - 1 customer) therefore highly limiting the opportunity of economy of scale for the provider, chopping off a bit chunk of the long tail. On the other hand, as described in Phil's post, an appliance can help overcome the data security concerns that many customers have as data will likely be stored within the appliance (inside the corporate firewall).
Another aspect I discussed a while back with a company interested in delivery their solution as an appliance was the reluctance from many Enterprise IT operators to punch a 'hole' into their firewall (allowing "Internet initiated" connections) required by that company to remote manage the appliance. Of course there are other ways of solving that problem typically by leveraging already existing VPN infrastructure etc. but this was going against the "just plug this appliance into your rack and everything works" (a.k.a. jptaiyraew) model.
Anyway, all these continuums and trends confirm something that we have believed in for some time now: the browser-only; cloud-only (a.k.a. BOCO) delivery of services is far from being the only way to deliver software a service and I would go as far as saying that it is actually a bit limitative view of SaaS.