So that confirms my point even more. That very rarely has there been a party's big win in a midterm election followed by a loss. And where I got the number 6 was from CandyCorn's post, which you seem to be lost on this.
You're changing your basis; it's not what you said before. But it's an interesting metric as well, so let's work it out.
Before 1870 midterm elections were all over the calendar, and the R party only came into being in 1854, so beginning with 1870:
1870 - (R POTUS): Rs lose seats but win 1872 POTUS. Theory holds.
1874 - (R POTUS): Rs lose seats, win 1876, but Ds win popular vote. Asterisk.
1878 - (R POTUS): Rs lose seats, win 1880 POTUS (by < 10,000 votes). Theory holds.
1882 - (R POTUS): Rs lose seats, lose 1884. Theory fails.
1886 - (D POTUS): Ds lose seats, lose 1888 but win popular vote. Asterisk.
1890 - (R POTUS): Rs lose seats, lose 1892. Theory fails. (third party (Weaver), no effect)
1894 - (D POTUS): Ds lose seats, lose 1896. Theory holds.
1898 - (R POTUS): Rs lose seats, win 1900. Theory holds.
1902 - (R POTUS): Rs lose seats, win 1904. Theory holds.
1906 - (R POTUS): Rs lose seats, win 1908. Theory holds.
1910 - (R POTUS): Rs lose seats, lose 1912. Theory fails. (third party (TR), no effect)
1914 - (D POTUS): Ds lose seats, win 1916. Theory holds.
1918 - (D POTUS): Ds lose seats, lose 1920. Theory fails.
1922 - (R POTUS): Rs lose seats, win 1924. Theory holds. (Third party (LaFollette), no effect)
1926 - (R POTUS): Rs lose seats, win 1928. Theory holds.
1930 - (R POTUS): Rs lose seats, lose 1932. Theory fails.
1934 - (D POTUS): Ds WIN seats, win 1936. Theory fails.
1938 - (D POTUS): Ds lose seats, win 1940. Theory holds.
1942 - (D POTUS): Ds lose seats, win 1944. Theory holds.
1946 - (D POTUS): Ds lose seats, win 1948. Theory holds (Significant Third party (Thurmond) cost Ds)
1950 - (D POTUS): Ds lose seats, lose 1952. Theory fails.
1954 - (R POTUS): Rs lose seats, win 1956. Theory holds.
1958 - (R POTUS): Rs lose seats, lose 1960 (by 120,000 votes). Theory fails. (Significant Third party (Byrd) cost Ds)
1962 - (D POTUS): Ds lose seats, win 1964. Theory holds.
1966 - (D POTUS): Ds lose seats, lose 1968 -- fails, but significant Third party (Wallace) cost Ds in close results: asterisk.
1970 - (R POTUS): Rs lose seats, win 1972. Theory holds.
1974 - (R POTUS): Rs lose seats, lose 1976. Theory fails.
1978 - (D POTUS): Ds lose seats, lose 1980. Theory fails. (Third party (Anderson) had no effect)
1982 - (R POTUS): Rs lose seats, win 1984. Theory holds.
1986 - (R POTUS): Rs lose seats, win 1988. Theory holds.
1990 - (R POTUS): Rs lose seats, lose 1992. Theory fails. (3rd party (Perot), no effect)
1994 - (D POTUS): Ds lose seats, win 1996. Theory holds. (Significant 3rd party (Perot), no effect)
1998 - (D POTUS): Ds WIN seats, lose 2000 (?) but Ds win popular vote. Asterisk ........ (3rd party (Nader) no effect)
2002 - (R POTUS): Rs WIN seats, win 2004. Theory fails.
2006 - (R POTUS): Rs lose seats, lose 2008. Theory fails.
2010 - (D POTUS): Ds lose seats, win 2012. Theory holds.
I make it 19 times "theory holds", 13 times "theory fails", plus four "asterisks". Even if you give all four asterisks to the "hold" column you've got 23-13. Go the other way, 19-17. I don't think you have a "very rarely" pattern there, or a pattern at all.