Which one is faster: one way reading ="single pass reading"

Hi,

Which one is faster: one way reading ="single pass reading"
Assumption :
a. Need to have 3 millions records
b. Need to have call 10  or 20 records repeatly
  (so for database it will be 10 times connection, each connection with one
record.
   or can be fancy 1 connection call return 10 sets of records)


1. Reading from Flat file
   Assume already give file name and just need to read the file
  (since it is flat file, each record represent a filename, with multiple
directory category)

2. Reading from XML file
   Assume schema already given just need to read the file
  (since it is xml file, each record represent an xml filename, with
multiple directory category)

3. Reading from Postgresql
   Assume primary key has been done with indexing
   just need to search the number and grap the text content
  (assume 3 millions of records, search the number, read the content file)

trying to recreate WebDBReader (from nutch) using C#
http://nutch.sourceforge.net/docs/api/net/nutch/db/WebDBReader.html

Thank you in advances,
Rosny