This is something the XML Schema group is going to have to decide. First of all, the client must be able to distinguish an interval of 1M as distinct from any arbitrary number of days. Though a set start date of the duration would allow it to be resolved to a fixed number of days, it may be used in to represent a repeating duration (you pay your rent on 1 Jan 2000 and then at 1 month intervals) where the number of days is not constant.
The issue for validation is how to do comparisons to allow the min and max constraints. I see two options: set a conversion factor for months and years that allow comparisions between durations where one of both terms contain fuzzy terms, or state that any comparision mixing years and months with days, hours, etc fail. If the first approach is used, I refined the values for months and years to their average length in a 400 year cycle in seconds in the Schema comment http://lists.w3.org/Archives/Public/www-xml-schema-comments/2000JanMar/0073. html This should be an improvement over 1Y == 360 days. If the second approach is used then the behavior would mimic the situation with NaN's in real numbers. (Also an issue that the Schema WG needs to directly resolve, though I think the discussion in http://lists.w3.org/Archives/Public/www-xml-schema-comments/2000JanMar/0025. html frame the expected behavior fairly well) Basically, any comparision between durations that involve both precise (days, hours, etc) and imprecise (months, years) would fail and violate the facet. This would result in some unexpected behavior. For example, PT1S would not be allowable if the max constraint was P10Y. Hopefully, you can design the timeDuration support so that either of these options could be implemented without radical surgery.
