query execution time

Hello

I've found a strange behavior in query execution:
I have really big table fti_author (>4 000 000 records) and i want to
find ID intersection of ones contain '^clarke' and '^arthur'

I've created 2 methods

1 select all with '^clarke' into temp table
select intersection of ones with '^arthur' and temp table
(example t1.txt in attachment)
2 one select with set of conditions
(example t2.txt in attachment)

THE PROBLEM is that 1st method requires 0.7sec and the 2nd - 3m20sec.
Please tell me why.

PS

Please find attached to this letter verbose plan for second query.

= selected dump ============================
--
-- TOC Entry ID 3 (OID 26649711)
--
-- Name: fti_author Type: TABLE Owner: gena
--
CREATE TABLE "fti_author" (
"string" character varying(50),
"id" oid
);

--
-- TOC Entry ID 7 (OID 26649711)
--
-- Name: "fti_author_s" Type: INDEX Owner: gena
--

CREATE INDEX "fti_author_s" on "fti_author" using btree ( "string" "varchar_ops" );

--
-- TOC Entry ID 8 (OID 26649711)
--
-- Name: "fti_author_i" Type: INDEX Owner: gena
--

CREATE INDEX "fti_author_i" on "fti_author" using btree ( "id" "oid_ops" );

====================

I'm using compiled CVS tree dated
23 Nov 2000.

complied with options
--enable-syslog --prefix= --exec-prefix=/usr --with-locale --with-perl

compiler
gcc version pgcc-2.91.66 19990314 (egcs-1.1.2 release)

System
Mandrake Linux 6.0 (with some updates)
Celeron 500/128M

Gena