Web API Library Cache Design Question

[Choonster TheMage <choonster.2010@...>]

I'm working on a library that provides an easy interface to a REST API
serving JSON[1]. The smallest results are about 1,000 characters
long[2] and the largest are about 684,000 characters long[3].

I'm caching the results so not every call to the methods result in a
full GET request being sent.

Currently I store the decoded table in the cache table and return a
shallow proxy table that has its __index pointing to the top level
result table. This prevents any modification of the top-level table,
but any tables nested in it are unprotected.

I can think of three different storage methods for the results to
prevent any modification of the cached value:

1) Use recursive proxies instead of shallow ones, i.e. access to a
nested table returns a proxy instead of the real table. This results
in a small overhead for each table lookup.

2) Deep copy the cached table and return the copy. This is simple, but
it may take a long time to copy the table version of link 3. This
results in a high initial overhead with raw table lookups for all
nested tables instead of the overhead of __index lookups.

3) Store the JSON string and decode it each time. This also has a high
initial overhead with raw table lookups (I'm not sure how much
overhead compared to the deep copy).

Does anyone have any opinions on these?

Regards,
Choonster

[1] http://blizzard.github.io/api-wow-docs/.
[2] https://gist.github.com/Choonster/5369708#file-itemdata-json
[3] https://gist.github.com/Choonster/5369708#file-auctiondata-json