One could get around the problem and still have unlicensed use, though, by time limiting the unlicensed authorization.  Just like build-out conditions on licenses, there would be a fixed time window within which widespread deployment should occur. If it doesn’t, the authorization is revoked.
This approach seems particularly relevant when an authorization holds great promise, but that promise is very uncertain, e.g. when the technology or the market is changing rapidly. “Sunsets” on rules are important since the passage of time invariably invalidates the premises of regulation, even as it entrenches the interests that coalesce around those regulations. 
The period of the authorization should be long enough to allow entrepreneurs to make a return on their investment, but short enough that an under-used authorization doesn’t lie fallow for decades. Fifteen years sounds about right. At the end of that period, the authorization would either (1) expire unless renewed, or (2) be automatically renewed unless canceled; even if renewed, the regulator could choose to make changes to operating parameters. I prefer option (1) since it would force proponents to make the case for the value of the authorization.
The criteria for renewal would be set at the time of initial authorization; they would measure the success of the authorization, and would probably include the installed base of devices and the growth rate. The renewal assessment should include not just the value of the use at that time, but also the potential value of an alternative use. Overall, the regulator would make a judgment about the success of its “investment” in this authorization just as a company or financier might judge a start-up they had funded; just like investors, the regulator is likely to be subject to confirmation bias, but at least it will be forced to revisit its decision in the light of new information.
The regulator would be required to make its determination some number of years (say five) before the authorization expires; if the authorization is not to be renewed, sales of equipment would have to cease at this point, e.g. five years before the authorization expires.
If the regulator decides that the unlicensed authorization has failed, it has a number of choices: it could retain the unlicensed designation but change the operating rules dramatically, it could redesignate the band for licensed use, or it could start thinking through the allocation of band again from scratch.
The two obvious historical test cases for such an approach would be the 2.4 GHz ISM allocation, where there was a meteoric take-off in use, and the 1920-1930 MHz Unlicensed PCS (UPCS) allocation, which has languished. Installed base numbers are always hard to find, but the FCC’s Equipment Authorization database should provide some indication, since it lists the devices that were authorized for sale. For example, UPCS saw 36 original grants over the period 3/20/1996 to 9/2/2011, whereas there were 358 grants in the 2400-2483.5 MHz ISM bands over the period 5/10/1994 to 8/11/2011. Investigation of the adoption rates in these two bands should provide a basis for choosing suitable allocation periods. It seems pretty clear from this chart that a decade is enough time to see whether an unlicensed authorization in a band is successful or not.
 Technically, users of Part 15 devices do not have any allocation status in the Commission’s rules; they are simply authorized to operate in terms of Part 15 rules. Instead, the Commission makes spectrum available for Part 15 devices on an unprotected and non-interference basis. See e.g. footnote 11 to para 5, Notice of Proposed Rulemaking in WT Docket No. 06-49, March 2006. I’ll use the terms “allocation” and “authorization” interchangeably here.
 See e.g. my “Internet Governance as Forestry: Deriving Policy Principles from Managed Complex Adaptive Systems” (2008), arguing that the more detailed a regulation, the more likely it is to become obsolete in the face of social, technological and commercial innovation; consequently, the more detailed the rules, the more rapidly regulations should expire (or be revisited).