From Microsoft's whitepaper explaining their vision (my emphasis below):
The Identity Metasystem is an interoperable architecture for digital identity that assumes people will have several digital identities based on multiple underlying technologies, implementations, and providers.
The metasystem enables identities provided by one identity system technology to be used within systems based on different technologies, provided an intermediary exists that understands both technologies and is willing and trusted to do the needed translations.
In ID-WSF, the 'technology' that is (currently) allowed to be different between the two providers, and that can be 'translated' to some extent, are security tokens. The web service provider (WSP) can, when it registers its service at the Discovery Service (DS), indicate which security token formats it expects (e.g. SAML, X.509, bearer, etc). When a client WSC) subsequently queries the DS for available services, it can also indicate which token formats it can support. It is the DS that looks for an intersection between the two different sets of security tokens and (acting as an STS) provides an appropriate token format to the WSC for inclusion in its subsequent request to the WSP.
Additionally, the DS may need to perform some translation between the security token that the WSC presents it as part of its discovery query and that which the DS returns to the WSC for inclusion in the request to the WSP. For instance, the WSC may present in its discovery query a SAML 1.1 assertion that it received from an IDP through ID-FF based SSO. If the relevant WSP (that being discovered by the WSC) only supports SAML 2.0 assertions, then the DS will have to translate from SAML 1.1 to SAML 2.0 to ensure that this WSP gets what it needs.
Admittedly, this level of flexibility on security token formats is not the wide-open freedom of different providers being able to choose different protocol stacks. But there is a price to be paid for that freedom.