For a complete reading of the ICP-3 document:
http://www.icann.org/icp/icp-3.htm

<...>
None of this precludes experimentation done in a manner that does not threaten the stability of name resolution in the authoritative DNS. Responsible experimentation is essential to the vitality of the Internet. Nor does it preclude the ultimate introduction of new architectures that may ultimately obviate the need for a unique, authoritative root. But the translation of experiments into production and the introduction of new architectures require community-based approaches, and are not compatible with individual efforts to gain proprietary advantage
<...>

5. Experimentation

Experimentation has always been an essential component of the Internet's vitality. Working within the system does not preclude experimentation, including experimentation with alternate DNS roots. But these activities must be done responsibly, in a manner that does not disrupt the ongoing activities of others and that is managed according to experimental protocols.

DNS experiments should be encouraged. Experiments, however, almost by definition have certain characteristics to avoid harm: (a) they are clearly labeled as experiments, (b) it is well understood that these experiments may end without establishing any prior claims on future directions, (c) they are appropriately coordinated within a community-based framework (such as the IETF), and (d) the experimenters commit to adapt to consensus-based standards when they emerge through the ICANN and other community-based processes. This is very different from launching commercial enterprises that lull users into a sense of permanence without any sense of the foregoing obligations or contingencies.

Moreover, it is essential that experimental operations involving alternate DNS roots be conducted in a controlled manner, so that they do not adversely affect those who have not consented to participate in them. Given the design of the DNS, and particularly the intermediate-host and cache poisoning issues described in Section 1 above, special care must be taken to insulate the DNS from the alternate root's effects. For example, alternate roots are commonly operated by large organizations within their private networks without harmful effects, since care is taken to prevent the flow of the alternate resource records onto the public Internet.

It should be noted that the original design of the DNS provides a facility for future extensions that accommodates the possibility of safely deploying multiple roots on the public Internet for experimental and other purposes. As noted in RFC 1034, the DNS includes a "class" tag on each resource record, which allows resource records of different classes to be distinguished even though they are commingled on the public Internet. For resource records within the authoritative root-server system, this class tag is set to "IN"; other values have been standardized for particular uses, including 255 possible values designated for "private use" that are particularly suited to experimentation.10

As described in a recent proposal within the IETF,11 this "class" facility allows an alternate DNS namespace to be operated from different root servers in a manner that does not interfere with the stable operation of the existing authoritative root-server system. To take advantage of this facility, it should be noted, requires the use of client or applications software developed for the alternate namespace (presumably deployed after responsible testing), rather than the existing software that has been developed to interoperate with the authoritative root. Those who operate alternate roots for global commercial purposes, however, have not followed this course.

In an ever-evolving Internet, ultimately there may be better architectures for getting the job done where the need for a single, authoritative root will not be an issue. But that is not the case today. And the transition to such an architecture, should it emerge, would require community-based approaches. In the interim, responsible experimentation should be encouraged, but it should not be done in a manner that affects those who do not consent after being informed of the character of the experiment.

Conclusion
<...>
ICANN – in deference to its public trust – will continue to collaborate with these citizens of the Internet community to advance the notions of a unique root system as a prerequisite to Internet stability, and to ensure that community-based policies take precedence. ICANN encourages responsible experimentation designed to further advance the Internet as a useful, stable, and accessible medium for the public good.