Sharded counter implementation not optimized for appengine pricing

[krullert]

The appengine sharded counter documentation
describes the implementation example as seen in: Sharded Counters Java Implementation

Whereas the implementation uses an unindexed property for the count.
I believe this causes a read-operation for each count entity, whereas
if I would make it indexed, and use a project query as;

final Query query = new Query(getKind())
    .addProjection(new PropertyProjection(CounterShard.COUNT, Long.class));

It would (according to the current quota/pricing page create a projection query

Projection queries that do not use the distinct on clause.
This type of query is counted as a single entity read for the query itself.
The individual results are counted as small operations.

So that would be a single entity read for the entire query, instead of readops being equal to the number of shards?

Is this correct or am I missing something?

See also: Quotas, App Engine Issue