While the original report outlines the issue, here's a solution proposal that is backwards compatible with the current spec:

In general:

• null either represents the null value or the null array (length -1)
• '[]' represents the empty array (length 0)
• '{}' (JSON object with no content) also represents the null value

With that, all options are covered for the Variant:
{"Type":12, "Body":{}} --> the null value (string of length -1)
{"Type":12, "Body": ""} --> the empty string (length 0)
{"Type":12, "Body": null} --> A null array (length -1)
{"Type":12, "Body": []} --> the empty array

The premise that the distinction between -1 and 0 length matters is wrong. See

5.1.9 Null, Empty and Zero-Length Arrays

A null array has no value. A zero-length or empty array is an array with 0 elements. Some DataEncodings will allow the distinction to be preserved on the wire, however, not all DevelopmentPlatforms will be able to preserve the distinction. For this reason, null, empty and zero length arrays are semantically the same for all DataEncodings.
...
Decoders shall be able to handle all variations supported by the DataEncoding, however, decoders are not required to preserve the distinction.
...

Agreed to no-fix in telecon.