• "A basic test for data file self-consistency, the absence of a checksum capability, prevents netCDF from being an acceptable archival format."
• "... it can be quite time consuming to read time series that are stored in a record variable in a large dataset."
• "Another issue is the limitation on variable size (even in the 64 bit offset variant) to ~4GB/record. I believe this will be problematic in the future as the size of variables (especially from model data) grows ..."
• "The netCDF convention is built to be very generic, which is great as a data format. However, as a community standard it may be overly broad. We need some additional convention/standard/guidelines on the netCDF file."
• "It is also easy for users to create NetCDF files without incorporating full metadata information inside the files. Operationally, this makes it difficult for archival and distribution."
• "In practice, netCDF depends on community conventions to be completely useful. For example, the use of coordinate variables is a convention, rather than being in the specification. I do think this is a weakness of the standard - newcomers will not infrequently create netCDF files thorough the API that are essentially `unusable' because they conform to the library API but not to the community conventions."

2009 Unidata NetCDF Workshop for Developers and Data Providers > Formats and Performance