Retrieving data from PostgreSQL to .NET application – performance test – surprising results

I did some performance tests (retrieving data from DB to .NET
application using Npgsql driver) and I found that for command type ==
StoredProcedure is better to first call function Prepare(). But, for
command type == Text (direct query) performance is better if I do not
run function Prepare(). Why it is so? I thought that Prepare() should
boost performance of executed query. Why executing stored procedure
with Prepare() is faster but direct query is slower?

Also I wonder why executing direct query is the fasted method for
retrieving data. Shouldn't stored procedures be faster?

Some details:
-10 000 iterations
-single connection for all iterations
-PostgreSQL on Windows
-query (the same in stored procedure)

SELECT count(*)
FROM "tableA" a
LEFT JOIN "tableB" b ON a.id = b."tableA_id"
LEFT JOIN "tableC" c ON b.id = c."tableB_id";

Results:
SP with Prepare() (Prepare for each iteration) 05.75s
SP with Prepare() (Prepare is called only for first iteration)
12.81s
SP without Prepare()
12.87s

Direct query with Prepare() (Prepare for each iteration):
06.73s
Direct query with Prepare() (Prepare is called only for first
iteration) 03.60s
Direct query without Prepare()
03.43s