Try BlockNews and/or UsenetNews again, an hour later.

You (DFS) replied ( to me ):

MessageID (instead of Article Number) "bookmarking" is tricky.
I list MessageIDs (100 at a time), starting from the highest Article Number,
until I find one of the 10 Message-ID "bookmarks" I've saved.

 
Can you give me the code you run to retrieve articles?
How were you getting 1000 articles in 4 seconds from glorb?

Glorb was lightning fast.
When possible, I sent it 100 commands at a time, instead of individually.

From "z1.CPP" in "Jeff-Relf.Me/z1.ZIP":

  typedef __int64  i64 ;   i64  _LowArtNum, _HighArtNum, HighArtNum ;

  const int  max_BookMarks = 10, MIDsPerPage = 100, Max_szMID = 200, Max_Sz_BookMark = max_BookMarks * Max_szMID ;

  OpenGroup₍₎( LnP NG ) {  //  211  303,341,573 1,844,605,951  2,147,947,523  Comp.OS.Linux.Advocacy
    if ( !Go( L"Group %s", LowerCaseNG( NG ) ) ) { Sh( L"No Such NewsGroup, “ %c%s%c ” ?", ChNew, NG, ChText );  return 0 ;  }
    strCpy( _T, *Ln.PP ), P = _T, gTok, gTok, gTok, LowArtNum = AtoI( Tok ) - 1, gTok, HighArtNum = AtoI( Tok );
    Sh( L"%s Articles;  From: %s  To: %s .", Commas( HighArtNum - LowArtNum ), Commas( LowArtNum + 1 ), Commas( HighArtNum ) ); return HighArtNum ;  }

  _LowArtNum = HighArtNum - MIDsPerPage - 1 ;
  Go( L"xHdr Message-ID %I64d-%I64d", _LowArtNum, _HighArtNum );

The python code I've relied on for years has become very slow and
unstable when running it against blocknews.  Connections are dropped all
the time.   SuckMT is OK fast, but it drops connections too.
 
I want to stay with blocknews because it has articles going back to
2003, and they might recover from their current woes.

Try BlockNews and/or UsenetNews again, an hour later.